Fusion: Towards Automated ICD Coding via Feature Compression

Fusion: Towards Automated ICD Coding via Feature Compression Junyu Luo Pennsylvania State University junyu@psu.edu Cao Xiao Amplitude danica.xiao@amplitude.com Lucas Glass IQVIA lucas.glass@iqvia.com Jimeng Sun UIUC jimeng@illinois.edu Fenglong Ma∗ Pennsylvania State University fenglong@psu.edu Abstract ICD coding aims to automatically assign Inter- national Classification of Diseases (ICD) codes from unstructured clinical notes or discharge summaries, which saves human labor and re- duces errors. Although several studies are pro- posed to solve this challenging task, none dis- tinguishes the importance of different phrases with a word window. Intuitively, informative phrases should be useful for the prediction. This paper proposes a feature compressed ICD coding model named Fusion to address this issue. In particular, we propose an attentive soft-pooling approach to compress the sparse and redundant word representations into infor- mative and dense ones as local features. Next, we use the key-query attention mechanism for modeling the inner relations among local fea- tures to generate the global features, which are further used to predict ICD codes. Experiments on two widely used datasets demonstrate that Fusion is comparable with baselines. We also find that none of the state-of-the-art approaches significantly perform better than others. Thus, automated ICD coding is still a challenging task. 1 Introduction The International Classification of Diseases (ICD) coding system helps standardize the recording of di- agnoses and treatments assigned to patients by med- ical professionals in the world. These ICD codes are generated from massive unstructured clinical notes. However, manual code assignments is labor- intensive and prone to errors. Thus, automatic ICD code assignment becomes an urgent need in the healthcare domain. Traditional machine learning methods (Larkey and Croft, 1996) tried to tackle this task based on feature extraction. However, it does not work well ∗Corresponding author. since clinical notes are noisy and complex. Re- cently, deep learning-based approaches (Cao et al., 2020; Xie et al., 2019; Li and Yu, 2020; Mullen- bach et al., 2018) are proposed to improve its perfor- mance. Among others, convolutional methods (Cao et al., 2020; Xie et al., 2019; Li and Yu, 2020; Mul- lenbach et al., 2018) outperform other approaches. Besides, some studies try to incorporate external in- formation to further improve the performance (Cao et al., 2020; Xie et al., 2019). However, they still suffer the following issues. • Redundant Information Deduction. The clin- ical notes are noisy and complex, where only some key phrases are highly related to the cod- ing. However, convolutional methods treat all the word windows equally, ignoring that differ- ent words have different importance and should be weighted differently within word windows. Besides, the sliding windows used in the con- volutional methods produce a lot of redundant information. Thus, it is important to reduce the non-informative and redundant information and distinguish the contributions of different convo- lutional features. • Interactions among Local Features. Most ex- isting approaches such as MultiResCNN (Li and Yu, 2020) only use the local features for coding obtained using different filters. However, they ignore the importance of interactions among dif- ferent local features. For …

Content truncated to 500 words